zephyr-7b-dpo-lora / eval_results.json
Jan Majkutewicz
End of training
5b9517b
raw
history blame
571 Bytes
{
"epoch": 1.0,
"eval_logits/chosen": -2.7746472358703613,
"eval_logits/rejected": -2.752934455871582,
"eval_logps/chosen": -310.119873046875,
"eval_logps/rejected": -321.6407165527344,
"eval_loss": 0.5893968343734741,
"eval_rewards/accuracies": 0.703499972820282,
"eval_rewards/chosen": -0.27382245659828186,
"eval_rewards/margins": 0.32820531725883484,
"eval_rewards/rejected": -0.6020277142524719,
"eval_runtime": 692.2285,
"eval_samples": 2000,
"eval_samples_per_second": 2.889,
"eval_steps_per_second": 0.361
}